Abstract
Much of today’s success in Information Retrieval (IR) comes from a hard approach: employing blazingly fast machines, ever more refined statistics, and increasingly powerful classification schemes. In recent years, however, the hard approach has entered a phase of diminishing returns. This paper explores a softer alternative which, we argue, is still in the phase of increasing returns. As the quality of an IR system is ultimately decided by its users, the approach starts from how these users structure information. Interestingly, for this approach many useful principles are readily available in the psychological literature. We illustrate the approach with three examples. The first applies the cognitive status of ‘complex nominals’ to improve search results by automatically constructing specialized queries. The second shows how the connection between language and imagery at the ‘basic level’ can be used for multimedia retrieval on the World Wide Web. The final example employs the notion of ’semantic space’ to make retrieval more effective especially for large scale corpora. In each example the results were substantial. The cases we studied illustrate how an approach to information retrieval based on cognitive principles can lead to significant, immediate, and fundamental results. It shows how prolific the application of cognitive science to the core of IR can be, and we believe that both disciplines stand to benefit from this approach.
